Tatyana V. Pestova, 1,2 Ivan N. Shatsky, 2 Simon P. Fletcher, 3 Richard J. Jackson, 3 and Christopher U.T. Hellen 1,4

Transcription

1 A prokaryotic-like mode of cytoplasmic eukaryotic ribosome binding to the initiation codon during internal translation initation of hepatitis C and classical swine fever virus RNAs Tatyana V. Pestova, 1,2 Ivan N. Shatsky, 2 Simon P. Fletcher, 3 Richard J. Jackson, 3 and Christopher U.T. Hellen 1,4 1 Department of Microbiology and Immunology, Morse Institute for Molecular Genetics, State University of New York Health Science Center at Brooklyn, Brooklyn, New York 11203; 2 A.N. Belozersky Institute of Physico-Chemical Biology, Moscow State University, Moscow, Russia; 3 Department of Biochemistry, University of Cambridge, Cambridge, CB2 1QW, UK Initiation of translation of hepatitis C virus and classical swine fever virus mrnas results from internal ribosomal entry. We reconstituted internal ribosomal entry in vitro from purified translation components and monitored assembly of 48S ribosomal preinitiation complexes by toe-printing. Ribosomal subunits (40S) formed stable binary complexes on both mrnas. The complex structure of these RNAs determined the correct positioning of the initiation codon in the ribosomal P site in binary complexes. Ribosomal binding and positioning on these mrnas did not require the initiation factors eif3, eif4a, eif4b, and eif4f and translation of these mrnas was not inhibited by a trans-dominant eif4a mutant. Addition of Met trna Met i, eif2, and GTP to these binary ribosomal complexes resulted in formation of 48S preinitiation complexes. The striking similarities between this eukaryotic initiation mechanism and the mechanism of translation initiation in prokaryotes are discussed. [Key Words: Hepatitis C virus; mrna; translation initiation; protein synthesis; ribosome] Received June 12, 1997; revised version accepted October 29, Protein synthesis begins following assembly of an initiation complex in which the initiation codon of an mrna and the anticodon of initiator trna are base-paired in the ribosomal P site. There are similarities and significant differences in the mechanisms of initiation complex formation in prokaryotes and eukaryotes. A universal characteristic is that initiation starts with separated ribosomal subunits. In prokaryotes, the small (30S) ribosomal subunit binds mrna and initiator trna in random order to form a complex that then undergoes conformational rearrangement, promoting codon anticodon base-pairing at the P site and joining of the large (50S) subunit (Gualerzi and Pon 1990). Ribosome binding results from interactions between the 30S subunit and multiple recognition elements in the mrna, such as the Shine-Dalgarno sequence (McCarthy and Brimacombe 1994). It does not depend on initiation factors. Ribosome-binding sites can 4 Corresponding author. chellen@netmail.hscbklyn.edu; FAX (718) occur at any position within an mrna and as a result many prokaryotic mrnas are polycistronic. In contrast, the small (40S) ribosomal subunit in eukaryotes requires several eukaryotic initiation factors (eifs), first to bind initiator trna (as a ternary complex with eif2 and GTP) to form a 43S complex and then to bind mrna to form a 48S complex (Merrick 1992). The most common mechanism for recruitment of an mrna is mediated by its capped 5 end, which is bound by the eif4e subunit of eif4f and then by the 43S complex. Ribosomal binding to mrna and scanning to the initiation codon require ATP hydrolysis and involve eif4a, eif4b, and eif4f. Most mrnas that use this mechanism of ribosomal binding are monocistronic because initiation is usually limited to the 5 -most AUG codon. A second, cap-independent mechanism of ribosome binding is used by mrnas whose 5 -untranslated region (5 UTR) contains an internal ribosomal entry segment (IRES). IRES-mediated initiation is used by several cellular mrnas that encode regulatory proteins, and has been usurped by numerous viruses (e.g., Pestova et al. 1996a). Two major groups of viral IRESs have been iden- GENES & DEVELOPMENT 12: by Cold Spring Harbor Laboratory Press ISSN /98 $5.00; 67

2 Pestova et al. tified; there is no obvious similarity between them or other viral or cellular IRESs. One of these groups is exemplified by the IRES of encephalomyocarditis virus (EMCV), which is 450 nucleotides long, highly structured, and lies immediately upstream of the initiation codon. The secondary and/or tertiary structure of the IRES is critical for its ability to promote initiation. We reconstituted EMCV IRES-mediated initiation in vitro from purified translation components and found that this process does not require noncanonical initiation factors. It is ATP-dependent and uses the same set of canonical eifs as cap-mediated initiation (Pestova et al. 1996a,b). It does not involve interaction between eif4e and an m 7 G cap. Instead, binding of the 43S complex to the EMCV IRES involves a specific cap-independent interaction between the 4G subunit of eif4f and the IRES. Hepatitis C virus (HCV) and pestiviruses such as classical swine fever virus (CSFV) and bovine viral diarrhea virus constitute another IRES group (Brown et al. 1992; Tsukiyama-Kohara et al. 1992; Poole et al. 1995; Wang et al. 1995; Rijnbrand et al. 1997). Ribosomal entry also occurs at or immediately upstream of the initiation codon on HCV-like IRESs and initiation probably also does not involve scanning (Reynolds et al. 1995, 1996; Rijnbrand et al. 1995, 1996, 1997; Wang et al. 1995; Honda et al. 1996; Lu and Wimmer 1996). Several properties, however, distinguish HCV-like IRESs from EMCV-like IRESs. They are 100 nucleotides shorter than EMCV-like IRESs and contain up to 30 nucleotides of coding sequence as well as functionally important structures, such as a pseudoknot upstream of the initiation codon (Fig. 1) that are unrelated to structural elements in the EMCV IRES. We report here that we have reconstituted HCV and CSFV IRES-mediated initation in vitro up to the stage of 48S complex formation using purified components (Met trna i Met, 40S subunits, initiation factors, and RNA) to identify which factors are requred for this process and begin to characterize their functions during it. Unexpectedly, the mechanism of initiation mediated by HCV and CSFV IRESs differed significantly from both cap- and EMCV IRES-mediated modes of initiation. Multiple interactions between 40S subunits and HCV and CSFV IRESs resulted in formation of stable binary complexes. The complex structure of these IRESs determined the correct positioning of the initiation codon in the ribosomal P site of binary complexes. Ribosomal binding to these mrnas and positioning on them did not require, and were not affected by, the initiation factors eif4a, eif4b, and eif4f. Complexes (48S) assembled in the absence of eif4a, eif4b, and eif4f were competent to form 80S complexes active in methionylpuromycin synthesis. Moreover, translation of HCV and CSFV mrnas differed from cap- and EMCV IRES-dependent mrnas in that it was not inhibited by a trans-dominant eif4a mutant. Binary ribosomal complexes required addition of only Met trna i Met, eif2, and GTP to form 48S complexes. These striking similarities between the mechanism of initiation on these viral IRESs and initiation on prokaryotic mrnas are discussed. Results HCV and CSFV RNAs bind 40S ribosomal subunits in the absence of initiation factors HCV RNA is translated efficiently in rabbit reticulocyte lysate (RRL) (Tsukiyama-Kohara et al. 1992). Assembly of 48S ribosomal preinitiation complexes on the HCV IRES was analyzed initially by incubating HCV (nucleotides ) RNA in RRL under normal translation conditions in the presence of GMP PNP and then resolving ribosomal and ribonucleoprotein (RNP) complexes by sucrose density gradient centrifugation. GMP PNP is a nonhydrolyzable GTP analog that causes 48S complexes to accumulate. Complexes (48S) formed under these conditions on up to 30% of the input [ 32 P]UTP-labeled HCV nucleotides RNA (Fig. 2A). The canonical factors eif2, eif3, eif4a, eif4b, and eif4f mediate attachment ( internal entry ) of 40S ribosomal subunits to the EMCV IRES (Pestova et al. 1996a). Purified 40S subunits also bound the HCV IRES in the presence of Met trna Met i, ATP, GMP PNP, and these five factors (Fig. 2A). Each factor and cofactor used to assemble this ribosomal complex was omitted from the reaction to determine whether it was essential. Any one, or even all of them, could be omitted without impairing binding of 40S subunits to the HCV IRES (Fig. 2B,C). In parallel reactions, 40S subunits also bound the CSFV IRES directly without factors or cofactors (Fig. 2H J) but did not bind the EMCV IRES or to -globin mrna under these conditions (Fig. 2D; data not shown). The 40S subunits were therefore not contaminated by initiation factors or by nonspecific RNA-binding proteins. HCV and CSFV IRESs have similar structures that are not related to the EMCV IRES (Fig. 1; Brown et al. 1992; Wang et al. 1995). Deletion of the initiation codon and coding region did not prevent 40S ribosomal subunits from binding to the HCV IRES (Fig. 2F). Neither HCV nor CSFV IRESs bound to active wheat germ 40S subunits (Fig. 1E; data not shown). These results indicate that rabbit 40S ribosomal subunits do not require initiation factors to bind HCV and CSFV IRESs, that their interaction with these RNAs is specific, and that it is stable enough to withstand sucrose density gradient centrifugation. Binary IRES 40S subunit complexes arrest primer extension within the pseudoknot and downstream of the initiation codon of HCV and of CSFV Primer extension inhibition ( toeprinting ) has been used to detect binary prokaryotic 30S ribosomal subunit- mrna complexes (Hartz et al. 1991). We used this method to analyze binding of mammalian 40S ribosomal subunits to HCV and CSFV IRESs. Toeprinting involves cdna synthesis by reverse transcriptase (RT) on a template RNA to which a ribosome or protein is bound. cdna synthesis is arrested either directly by the bound complex, yielding a stop or toeprint at its leading edge, or indirectly, by stabilization of adjacent helices (Hartz et al. 1988; Baker and Draper 1995). The resulting toeprints are located on a sequencing gel. Toeprinting is a more 68 GENES & DEVELOPMENT

3 Internal translation initiation Figure 1. Model secondary and tertiary RNA structures of the 5 NTRs of hepatitis C virus (A) and classical swine fever virus (B), based on proposals by Brown et al. (1992), Honda et al. (1996), and Wang et al. (1995). The nomenclature of domains is as described by Honda et al. (1996). HCV constructs were linked either to NS or CAT reporter cistrons both sequences are shown. An NS reporter cistron (not shown) was linked 74 nucleotides downstream of the CSFV initiation codon in all CSFV constructs. HCV and CSFV initiation codons are underlined. Stop-sites where primer extension was arrested on naked RNA only are indicated by open circles; sites of RT arrest caused or enhanced by binding of eif3 and of 40S subunits are indicated by solid diamonds and circles, respectively. The 3 border of the HCV nucleotides deletion and both borders of the HCV nucleotides and HCV nucleotides deletions are labeled and indicated by asterisks. stringent assay of RNA protein interaction than sucrose density gradient centrifugation. For example, cytoplasmic RNA-binding proteins (including initiation factors) form RNP complexes readily on capped eukaryotic mrnas but do not arrest primer extension (Anthony and Merrick 1992). In all toeprinting experiments described here (except Fig. 8B, below), cdna products contained a single radioactive moiety derived from the endlabeled primer. The intensity of a toeprint is therefore directly proportional to the frequency of arrest at a specific position. The positions of all the RT stops described below are shown on the structural models of HCV and CSFV IRESs in Figure 1. The HCV IRES is highly structured (Fig. 1A) and as a GENES & DEVELOPMENT 69

4 Pestova et al. Figure 2. Ribosomal complex formation on HCV and CSFV IRESs. Assays were done using HCV nucleotides RNA (A F) and HCV, CSFV, and EMCV RNAs (D,F J) as indicated and with RRL or rabbit 40S subunits and eif2, eif3, eif4a, eif4b, and eif4f (A) with rabbit 40S subunits and combinations of eif2, eif3, eif4a, eif4b, and eif4f (B,C) and with rabbit or (where indicated) wheat germ 40S subunits (D J). ATP was included only in reactions that contained eif4a, eif4b, or eif4f; Met Met trna i and GMP PNP were included only in reactions that contained eif2. Other components described in Materials and Methods were present in all reactions. Sedimentation was from right to left. The positions of ribosomal complexes are indicated by arrows. (G) HCV IIa CAT, IIIb CAT, and IIIc CAT mrnas are identical to wild-type HCV CAT mrna except for deletions of HCV nucleotides 28 67, , and , respectively. Sucrose gradients shown in H were centrifuged for a shorter time than others shown here. Fractions from upper parts of the sucrose gradient have been omitted from graphs in H for greater clarity. result, RT arrest occurs at a number of sites on naked HCV RNA. In the experiments described here, primer extension on RNA that consisted of HCV nucleotides linked to a truncated influenza NS cistron (NS ) was arrested at several stable structures in the IRES, yielding strong stops at positions that included G 318, G 320,U 324, and U 329 in the pseudoknot (Fig. 3A, lane 1). Toeprint analysis of binary HCV RNA 40S subunit complexes indicated that 40S subunits enhanced RT stops strongly at G 318 and G 320 in the pseudoknot and arrested primer extension at C 355 and to a lesser extent at U 356, downstream of the initiation codon AUG (Fig. 3A, lane 2). Primer extension on naked CSFV (nucleotides 1 442) NS RNA was arrested strongly at U 304 near the base of domain III but not significantly in the pseudoknot (Fig. 3D, lane 1). Analysis of binary CSFV RNA 40S subunit complexes indicated that 40S ribosomal subunits arrested primer extension strongly at C 334 and to a lesser extent at G 345 in the pseudoknot, as well as strongly at U 388 and to a lesser extent at U 387 and U 389 downstream of the initiation codon AUG (Fig. 3D, lane 2). Binding of 40S subunits at these contact points on HCV and CSFV IRESs might prevent detection of additional IRES 40S subunit contacts using the same primers, for example, on the opposite side of the pseudoknot. To address this possibility, and the possibility that additional 40S subunits might bind elsewhere on these IRESs, we carried out additional toeprint analyses using new primers that were chosen to bind upstream of the 5 -most stops described above. The results of these additional experiments are shown in Figure 3, B and E. These panels show that no additional contacts between 40S subunits and these IRESs were identified other than 70 GENES & DEVELOPMENT

5 Internal translation initiation Figure 3. Toeprint analysis of 48S complex formation on HCV and CSFV IRESs. (A) Wild-type HCV (nucleotides ) NS RNA was incubated with 40S ribosomal subunits (lanes 2 6), eif2 (lanes 3 6), Met trna Met i (lanes 3 5), eif3 (lanes 4 6), and eif4a, eif4b, and eif4f (lanes 5,6); (B) wild-type HCV (nucleotides ) NS RNA was incubated with 40S ribosomal subunits (lanes 2 4), eif2, Met trna Met i, and GMP PNP (lanes 3,4), eif3 (lanes 4,5), and eif4a, eif4b, and eif4f (lane 4); (C) wildtype HCV (nucleotides ) NS (lanes 1 3), (AUG AAG) mutant HCV (nucleotides ) NS (lanes 4 6), and (AUG GCG) mutant HCV (nucleotides ) NS RNAs (lanes 7 9) were incubated with 40S ribosomal subunits (lanes 2,3,5,6,8,9) and with eif2 and eif3, Met trna Met i and GMP PNP (lanes 3,6,9); (D) wild-type CSFV (nucleotides 1 442) NS RNA was incubated with 40S ribosomal subunits (lanes 2 6), eif2 (lanes 3 6), Met trna i Met (lanes 4 6), eif3 (lanes 5 6), and eif4a, eif4b, and eif4f (lane 6); (E) wild-type CSFV (nucleotides 1 442) NS RNA was incubated with 40S ribosomal subunits (lanes 2 5), eif2, Met trna imet and GMP PNP (lanes 3 5), eif3 (lanes 4 6), and eif4a, eif4b, and eif4f (lane 5). Reaction conditions are described in Materials and Methods. In A,C, and D, the primers (5 -GG- GATTTCTGATCTCGGCG-3 ) and (5 -CTC- GTTTGCGGACATGCC-3 ) were annealed to the NS cistron 130 nucleotides downstream of the HCV initiation codon and 110 nucleotides downstream of the CSFV initiation codon, respectively, and were extended with AMV RT. In B and D, the primers 5 -CGCAAGCACCCTATC-3 (complementary to HCV nucleotides ) and 5 -CCTGATAGGGTGCTGCAG-3 (complementary to CSFV nucleotides ) were annealed to HCV and CSFV IRESs, as appropriate, and were extended with AMV RT. Full-length cdna is marked E. Other cdna products terminated at the sites are indicated on the right. Reference lanes C,T,A, and G depict HCV or CSFV sequences, as appropriate. GENES & DEVELOPMENT 71

6 Pestova et al. those that were described above. Subunits (40S) do not, therefore, bind anywhere else on these IRES elements other than at the position described above, including the opposite side of the HCV and CSFV pseudoknots and at or near to the 5 terminii of these RNAs. These results are summarized in Figure 1. They confirm that 40S ribosomal subunits alone can bind directly to HCV and CSFV IRESs, probably at or close to the pseudoknot. The observation that binary complexes yielded toeprints downstream of HCV and CSFV initiation codons indicate that the coding regions of both RNAs are fixed stably in the mrna-binding cleft of 40S subunits. This observation may also provide an explanation for the importance of the conserved sequences adjacent to the initiation codon for IRES function (Reynolds et al. 1995; Honda et al. 1996; Lu and Wimmer 1996). The 43S ribosomal complex binds by itself at the initiation codon of HCV and CSFV IRESs without eif4a, eif4b, or eif4f Eukaryotic translation initiation is analyzed conventionally using sucrose density gradient centrifugation to resolve RNP, preinitiation, and initiation complexes. This approach is not appropriate for analysis of initiation on HCV and CSFV IRESs because as the results described above have shown, 40S subunits bind directly to these RNAs without aminoacylated initiator trna or initiation factors. Primer extension analysis, however, is a sensitive assay of prokaryotic initiation (Hartz et al. 1988) and has been used to assay cap- and IRES-mediated modes of eukaryotic initiation (Anthony and Merrick 1992; Pestova et al. 1996a,b). This method was used to dissect the initiation process on HCV and CSFV IRESs. Toeprint analysis of complexes assembled from eif2, 40S subunits, Met trna Met i and GMP PNP on HCV (nucleotides ) NS RNA yielded prominent stops at AAA (Fig. 3A, lane 3). These stops were not observed during toeprint analysis of complexes assembled from the ternary [eif2/initiator trna/gmp PNP] complex and HCV (nucleotides ) NS mrna in the absence of 40S subunits (data not shown). This result and the position of these stops relative to the initiation codon indicates that they are caused by a stable ribosomal complex assembled at the initiation codon. The stops that had been detected at C 355 and U 356 in the binary HCV IRES 40S complex were not apparent and the G 318 and G 320 stops were weaker (Fig. 3A, lane 2). The same stops at G 318,G 320, and AAA were detected on HCV RNA when eif3 or eif3, eif4a, eif4b, and eif4f (Fig. 3A, lanes 4,5) were included in assembly reactions in addition to eif2, Met trna Met i GMP PNP and 40S subunits. Therefore eif3, eif4a, eif4b, and eif4f did not alter the interaction of the 40S subunit with the HCV IRES. We have shown previously that these factors are absolutely required for assembly of 48S complexes on the EMCV IRES (Pestova et al. 1996a,b). We therefore used 48S complex formation on the EMCV IRES to confirm that the factors used in reconstituting 48S complex formation on the HCV and CSFV IRESs were all active (data not shown). A band at A 243 became more prominent when eif3 was present in reactions (Fig. 3A, lanes 4 6). Closer inspection revealed that this band consisted of stops at A 243 and A 244 (Fig. 3B, lanes 4,5). The stops at CU were apparent, whereas those at AAA were not when Met trna Met i was not included in a reaction that contained eif2, eif3, GMP PNP, and 40S subunits (Fig. 3A, lane 6). This result suggests that RT arrest at AAA requires cognate codon anticodon base-pairing between HCV mrna and initiator trna. To address this question, 48S complex assembly was analyzed on AUG AAG and AUG GCG mutant IRESs that differ from wild type in the initiation codon. These triplets are not active initiation codons in these RNAs (Reynolds et al. 1995). The efficiency of binding of 40S subunits to these wild-type and mutant RNAs was indistinguishable by sucrose density gradient centrifugation (data not shown). The strong stops at AAA that are characteristic of 48S complex formation at the initiation codon, however, were detected only on wildtype HCV RNA and not on either mutant (Fig. 3C, lanes 3,6,9). These observations strongly support the conclusion that codon anticodon base-pairing is required for stable IRES-mediated assembly of a 48S complex at the HCV initiation codon. We next used toeprinting to analyze ribosomal complex formation on the CSFV IRES. Toeprint analysis of complexes assembled from 40S subunits, eif2, Met trna Met i, and GMP PNP on CSFV (nucleotides 1 442) NS RNA yielded stops at C 334 and G 345 in the pseudoknot that were of similar intensity, and stops at UUU that were weaker than for binary CSFV IRES 40S complexes (Fig. 3D, lane 2,4). In addition, very prominent stops appeared at UGA (Fig. 3D, lane 4) that were not detected on these binary complexes (Fig. 3D, lane 2). These stops were not detected on analysis of complexes assembled from eif2, GMP PNP, and 40S subunits without Met trna Met i (Fig. 3D, lane 3). The position of these stops relative to the initiation codon indicates that 43S complexes alone bind correctly to the CSFV initiation codon AUG , arresting primer extension at UGA The stops derived from complexes that had assembled on the CSFV IRES from eif2, Met trna Met i, GMP PNP, and 40S subunits were similar to those derived from these components and eif3 (Fig. 3D, lanes 4,5) except that in the latter instance (1) a faint stop was detected at A 250 and the stop at U 304 was stronger and (2) the stops at UUU were relatively weaker and those at UGA were relatively stronger. The first of these results suggest that eif3 binds specifically to the CSFV IRES. Closer inspection indicated that eif3 arrested primer extension at A 250 and C 251 (Fig. 3E, lanes 4 6). The second of these results indicates that formation of 48S complexes is enhanced in the presence of eif3 vis-á-vis formation of binary IRES 40S subunit complexes, and therefore suggests that a role of eif3 in CSFV initiation may be to stabilize the 48S complex on the initiation codon. Neither the pattern nor 72 GENES & DEVELOPMENT

7 Internal translation initiation the intensity of stops that were detected on CSFV RNA were altered as a result of inclusion of eif4a, eif4b, and eif4f in reactions that contained eif2, eif3, Met trna i Met, GMP PNP, and 40S subunits (Fig. 3D, cf. lanes 5 and 6). Therefore, these three factors do not affect the interaction of the 43S complex with the CSFV IRES. In a further series of experiments, toeprinting analysis of complexes assembled on HCV and CSFV IRESs in reactions that contained eif2, eif3, Met trna i Met, GMP PNP, and 40S subunits with or without eif4a, eif4b, and eif4f using a different set of primers revealed that 48S complexes did not assemble either at or near to the 5 ends of these mrnas or at or near to any AUG codon other than the authentic initiation codon (Fig. 3B,E). Taken together, these results indicate that a 43S ribosomal preinitiation complex is able, by itself, to bind directly and stably to the authentic initiation codon on HCV and CSFV IRESs without the involvement of noncanonical factors, or of eif4a, eif4b, or eif4f. Stable binding of this complex at the initiation codon requires initiator trna. It may be enhanced by, but does not require, eif3, which is present in stoichiometric amounts on native 40S subunits (Sundkvist and Staehelin 1975). The 40S subunit itself, however, is responsible for the specificity of binding of the 43S complex to these IRESs. eif3 alone binds specifically to HCV and CSFV IRESs Inclusion of eif3 in reactions that contained eif2 and 40S subunits resulted in RT arrest at AC and U 304 on the CSFV IRES and at AA on the HCV IRES (Fig. 3, A, lanes 4 6; B, lanes 4,5; D, lanes 5,6; E, lanes 4 6). Appropriate RNP complexes were toeprinted to determine whether eif3 alone binds specifically to these RNAs. eif3 alone arrested RT at these positions on the CSFV IRES (Fig. 3E, lane 6 and Fig. 4A, lane 2) and on the HCV IRES (Fig. 4B, lane 2). These results indicate that eif3 binds specifically to domain III of HCV and CSFV IRES elements. The specificity of both of these interactions has been confirmed by chemical and enzymatic footprinting of binary eif3 IRES complexes (D. Sizova, V. Kolupaeva, T. Pestova, I.N. Shatsky, and C. Hellen, in prep.). Factor requirements for assembly of active 80S complexes on HCV and CSFV IRESs Complexes (48S) assemble accurately at the initiation codon on HCV and CSFV IRESs on inclusion of eif2, initiator trna, and GMP PNP (or GTP) with 40S subunits. To determine whether these 48S complexes are bona fide intermediates in the initiation process that culminates in formation of the first peptide bond, we assayed the ability of these complexes to join with 60S ribosomal subunits to form 80S complexes and for these 80S complexes to catalyze methionylpuromycin synthesis. This model reaction mimics dipeptide formation. Assembly of active 80S complexes was mediated by a ribosomal salt wash (RSW) fraction (the M KCl Figure 4. Primer extension analysis of RNP complexes formed on CSFV and HCV IRES elements as follows. (A) CSFV (nucleotides 1 442) NS and (B) HCV (nucleotides ) NS RNAs were incubated with (lane 2) or without eif3 (lane 1) under standard conditions. Primers were annealed to the NS -coding sequences of these RNAs and extended with AMV RT. The full-length cdna extension product is marked E. The cdna products labeled A 250,U 304, and A 243 on the right terminated at these nucleotides. The reference lanes C, T, A, and G depict the CSFV sequence (A) and the HCV sequence (B). Positions of CSFV and HCV nucleotides are indicated on the left of the appropriate panel. phosphocellulose elution fraction derived from the 50% 70% A.S. cut) that contained eif5 and eif5a (which are required for subunit joining and peptide bond formation) but that did not contain eif4a, eif4b, or eif4f (Benne et al. 1978). The absence of eif4a, eif4b, and eif4f from this fraction was confirmed by Western blotting (data not shown). Toeprinting assays showed that this RSW fraction had no effect on the amount of 48S complex formed on HCV and CSFV IRESs or on the position of these complexes (data not shown). There was no requirement for eif4a, eif4b, or eif4f in formation of 80S complexes on the CSFV IRES, but eif3 was absolutely essential for this process (Fig. 5A). Moreover, omission of eif3 from the 80S assembly reaction also caused the destabilization of 48S complexes (Fig. 5A). Formation of 80S complexes was accompanied by a decrease in the amount of 48S complexes. The formation of 80S complexes was not observed if the RSW fraction or 60S subunits were omitted from assembly reactions, or if GTP was substituted by GMP PNP (Fig. 5A; data not shown). Similarly, eif4a, eif4b, and eif4f were not required for assembly of 80S complexes capable of methionylpuromycin synthesis (Fig. 5C). As expected, and consistent with previous reports (Trachsel et al. 1977; Benne and Hershey 1978; Grifo et al. 1982), eif4a, eif4b, and eif4f were essential for assembly of 80S complexes on -globin mrna that were competent to catalyze methionylpuromycin synthesis (Fig. 5B,C). This result serves as an additional control that the RSW fraction used to mediate subunit joining was not contaminated GENES & DEVELOPMENT 73

8 Pestova et al. determine whether eif3 is involved in altering the conformation of the 48S complex so that it can bind to 60S subunits or whether it is involved more directly in joining of subunits. Figure 5. Assembly of 80S ribosomal initiation complexes on the CSFV IRES and on -globin mrna and analysis of their activity in stimulating methionylpuromycin synthesis. Assays were done using CSFV (nucleotides 1 442) NS mrna (A,C) and -globin mrna (B,C) and with initiation factors and other translation components as described in Materials and Methods. (A,B) Sedimentation was from right to left, and the formation of complexes was assayed by incorporation of [ 35 S]methionine trna. The positions of ribosomal complexes are indicated. Fractions from upper parts of sucrose gradients have been omitted from graphs in A and B for greater clarity. (C) The methionylpuromycin synthesis assay was done as described in Materials and Methods, using -globin mrna (columns 1 and 2); CSFV (nucleotides 1 442) NS mrna (columns 3 5); eif2, eif3, eif4a, eif4b, eif4f (columns 1 and 3), eif2 and eif3 (columns 2 and 4); and eif2, eif4a, eif4b, and eif4f (column 5). All reactions also contained GTP, 40S and 60S ribosomal subunits, and a 50% 70% A.S. RSW subfraction as described in Materials and Methods. A background value determined with charged initiator trna alone was subtracted from all panels. Translation of HCV and CSFV mrnas is resistant to inhibition by trans-dominant eif4a(r362q) Cap- and EMCV IRES-mediated initiation both require eif4a, probably as a subunit of eif4f (Trachsel et al. 1977; Benne and Hershey 1978; Grifo et al. 1983; Pestova et al. 1996b). Both processes are inhibited by the transdominant eif4a (R362Q) mutant; inhibition is relieved by eif4f, and less effectively, by eif4a (Pause et al. 1994). The data above indicate that 43S ribosomal complexes bind to HCV and CSFV IRESs without eif4a, eif4b, or eif4f and can then bind 60S ribosomal subunits to form active 80S complexes. Translation of these mrnas would therefore be expected to be unaffected by eif4a(r362q). Addition of eif4a(r362q) to RRL programmed with dicistronic cyclin HCV IRES NS or cyclin CSFV IRES NS mrnas strongly impaired translation of the upstream cyclin cistron in both instances but had no effect on IRES-mediated translation of the downstream NS cistron (Fig. 6, lanes 3,6). Addition of wildtype eif4a to RRL did not influence translation of these mrnas (Fig. 6, lanes 2,5). The activity of the transdominant eif4a(r362q) mutant was confirmed by its ability to inhibit translation of luciferase mrna and of a GUS reporter cistron linked to the EMCV IRES (data not shown). These results indicate that eif4a (and probably eif4f) are not required for HCV/CSFV translation. They strongly support the results obtained by reconstituting initiation reactions from purified components. They are consistent with a model for initiation in which 43S complexes bind directly to these IRESs without the involvement of eif4a, eif4b, or eif4f. by eif4a, eif4b, or eif4f. As described for CSFV, formation of 80S complexes was accompanied by a decrease in the amount of 48S complexes. These results confirm that assembly of 48S preinitiation complexes on HCV and CSFV IRESs that are competent to complete all remaining stages in the initiation process does not require eif4a, eif4b, or eif4f. This process therefore differs fundamentally from cap-dependent initiation on -globin. Toeprinting assays showed that eif3 did not influence the efficiency or the fidelity of assembly of 48S complexes at the initiation codon of HCV and CSFV IRESs (Fig. 3A,D), but this factor was required for assembly of 80S complexes (Fig. 5A). Our data are not sufficient to Figure 6. Dominant-negative effect of the eif4a R362Q mutant protein on cap-mediated translation and its lack of effect on HCV and CSFV IRES-mediated translation. Reticulocyte lysate (10 µl) was preincubated alone (lanes 1,4), with eif4a wild-type (1.2 µg) (lanes 2,5), or eif4a mutant (1.2 µg) (lanes 3,6) for 5 min at 30 C, then incubated for 60 min at 30 C with dicistronic mrnas (0.5 µg) as described in Materials and Methods. Translation products were analyzed by autoradiography after electrophoresis on SDS 17% polyacrylamide gel. 74 GENES & DEVELOPMENT

9 Internal translation initiation Structural integrity of the HCV IRES is required for stable binding of 43S complexes at the initiation codon Disruption of HCV and CSFV IRES structure severely reduces their activity and must therefore impair one or more steps in the process of internal ribosomal entry. Initiation was analyzed on various mutant IRESs to identify these steps. The IIa, IIIb, and IIIc hairpins contribute to HCV IRES function (Rijnbrand et al. 1995). We examined their role in binding 40S ribosomal subunits and 43S preinitiation complexes to the IRES using wild-type HCV (nucleotides 1 349) CAT mrna and nucleotides 26 67, nucleotides , and nucleotides mutant derivatives, thereof, that lack IIa, IIIb, and IIIc hairpins, respectively (Fig. 1A). The results of sucrose density gradient centrifugation showed that these three hairpins are not essential for binary complex formation. Appropriate HCV deletion mutants all bound 40S subunits strongly (Fig. 2G). Toeprint analysis of binary ribosomal complexes assembled on wild-type HCV (nucleotides 1 349) CAT mrna yielded stops (Fig. 7A, lane 2) that differed slightly from those obtained on wild-type HCV (nucleotides ) NS mrna (Fig. 3A, lane 2). Bound 40S subunits enhanced RT stops strongly at G 318 and G 320 as described above, but also arrested primer extension at CAUGA (i.e., the HCV initiation codon) and at a GAG triplet nucleotides downstream of the initiation codon. These results are summarized in Figure Figure 7. Binding of 40S subunits and assembly of 48S complexes on HCV and CSFV mutant IRESs. (A) Wild-type HCV (nucleotides 1 349) CAT (lanes 1 3) and mutant HCV (nucleotides ) CAT (lanes 4 6) were incubated under standard conditions with 40S ribosomal subunits (lanes 2,3 and 5,6) and with eif2, eif3, Met trna Met i and GMP PNP (lanes 3,6). (B) Mutant HCV (nucleotides ) CAT (lanes 1 3), (nucleotides ) CAT (lanes 4 6) and (nucleotides ) (lanes 7 9) RNAs were incubated with 40S subunits (lanes 2,3,5,6,8,9) and with eif2, eif3, Met trna Met i, and GMP PNP (lanes 3,6,9). Reaction conditions are described in Materials and Methods. A primer (5 -GCAACTGACTGAAATGCC-3 ) was annealed to the CAT cistron 80 nucleotides downstream of the HCV initiation codon and was extended with AMV RT. Primer extension was arrested at sites indicated on the right either as nucleotides in the HCV IRES or as positions relative to the A of the HCV initiation codon. Reference lanes C, T, A, and G depict the HCV CAT sequence. The junction between the HCV IRES and the CAT cistron is indicated. This gel was exposed to x-ray film longer than that shown in A to show the differences in the intensity of stops on various mutant RNAs. (C) Model of the CSFV pseudoknot and adjacent nucleotides showing the substitutions present in the mutants UC 325,GA 357, and UC 325 GA 357 and the deletions made in the mutants A 145 U 148 and A 349 A 353. In this model, helix I of the pseudoknot comprises nucleotides and , and helix II comprises nucleotides and (D) CSFV (nucleotides 1 442) NS RNAs containing the mutations UC 325 (lanes 1 3), GA 357 (lanes 4 6), UC 325 GA 357 (lanes 7 9), and A 349 A 353 (lanes 10 12) were incubated with 40S subunits (lanes 2,3,5,6,8,9,11,12) and with eif2, eif3, Met trna Met i and GMP PNP (lanes 3,6,9,12). GENES & DEVELOPMENT 75

10 Pestova et al. 1A. They show that the interaction of nucleotides downstream of the pseudoknot with 40S ribosomal subunits is influenced by the sequence of the coding region. Toeprinting of binary complexes assembled on nucleotides mutant HCV CAT mrna yielded stops at CAUGA and at GAG ( nucleotides) of the same intensity as on wild-type RNA (Fig. 7A, lanes 2,5); on the nucleotides HCV mutant, corresponding stops were weaker (Fig. 7B, lane 8). The intensity of stops at G 318 and G 320 on these two mutant RNAs did not increase in the presence of 40S subunits. Toeprint analysis of binary complexes assembled on nucleotides HCV CAT RNA yielded strong stops at G 318 and G 320 and very weak stops both at CAUGA and at GAG ( nucleotides) (Fig. 7B, lane 2). The results of sucrose density gradient analysis indicate that the IIa, IIIb, and IIIc hairpins do not determine binding of 40S subunits to the HCV IRES; these findings are supported by the results of toeprint analysis. toeprinting also shows that these three hairpins are necessary for the wild-type pattern of interactions between the HCV IRES and the 40S subunit to occur. Toeprint analysis of 48S complexes assembled from eif2, eif3, Met trna Met i, GMP PNP, and 40S subunits on wild-type HCV (nucleotides 1 349) CAT yielded stops at CAUGA that were slightly weaker, stops at G 318 and G 320 that were similar in intensity and stops at A 243 and at GAGAA ( nucleotides, downstream of the A of the initiation codon) that were significantly stronger than those detected on analysis of the corresponding binary IRES 40S complex (Fig. 7A, lane 3). The prominent stops at GAGAA ( nucleotides) Met were not detected when Met trna i was omitted from reactions (data not shown) and were therefore caused by 48S complexes assembled at the initiation codon. These results show that the HCV IRES undergoes a conformational rearrangement when the ternary eif2/ GTP/initiator trna complex binds to the binary IRES 40S complex to form a 48S preinitiation complex. The stop at A 243 was attributable to bound eif3. Under similar conditions, this stop was not detected on nucleotides HCV RNA, was very weak on nucleotides HCV RNA and had near-wild-type intensity on nucleotides mutant RNA (Fig. 7B, lanes 3,6,9). These results suggest that binding of eif3 to this IRES requires hairpin IIIb and is enhanced strongly by hairpin IIIc. The stops at GAGAA ( nucleotides) caused by assembly of 48S complexes were much weaker on nucleotides mutant RNA than on wild-type HCV RNA (Fig. 7A, lanes 3,6). These stops were even weaker on nucleotides and nucleotides mutant RNAs than on nucleotides mutant RNA (Fig. 7B, cf. lanes 3,6, and 9). The gel in Figure 7B that shows these data was exposed to film longer than that in Figure 7A to show the differences in intensity of stops at CAUGA and at GAGAA ( nucleotides) between these different mutants (Fig. 7, cf. lanes 4 6 in A and B). These results show that deletion of IIa, IIIb, or IIIc hairpins from the HCV IRES strongly impairs assembly of 48S complexes at the initiation codon. They are consistent with observations that translation of HCV mrna is strongly impaired by deletion of any one of these hairpins (Rijnbrand et al. 1995). Sucrose density gradient analysis showed that binary ribosomal complexes assembled efficiently on these three mutant RNAs. The detailed toeprinting analysis presented here showed that despite efficient binary complex formation, the normal pattern of interaction between 40S subunits and the HCV IRES in the vicinity of the initiation codon was altered by these deletions. The result of this defect is that assembly of 48S complexes on the initiation codon was strongly impaired. Correct positioning of the 43S complex on the HCV IRES to form a 48S complex is more sensitive than binary complex formation to these structural changes. Correct positioning of the initiation codon in the ribosomal P site is therefore determined by RNA sequences within different regions of the IRES and is impaired by mutation of these sequences. Structural integrity of the CSFV pseudoknot is required for stable binding of 43S complexes at the initiation codon CSFV and HCV IRES elements both contain an essential pseudoknot nucleotides upstream of the intiation codon (Wang et al. 1995; Rijnbrand et al. 1997; S.P. Fletcher and R.J. Jackson, in prep.). We examined its contribution to CSFV IRES function using three mutants, UC 325,GA 357, and UC 325 GA 357, that have substitutions in helix II of the pseudoknot and one mutant, A 349 A 353, that has a deletion in the loop connecting helices I and II. Helix I is restored in the UC 325 GA 357 pseudorevertant. It has near-wild-type IRES activity, whereas that of all other mutants is strongly impaired (S.P. Fletcher and R.J. Jackson, in prep.). These mutations are illustrated in Figure 7C. Subunits (40S) bound similar proportions of wild type and of these four mutant CSFV RNAs (Fig. 2H). Primer extension was arrested strongly by 40S subunits at C 334 and G 345 on all four mutants (Fig. 7D, lanes 2,5,8,11). The stops at UUU were much weaker on the UC 325,GA 357, and A 349 A 353 mutant RNAs that had been incubated with 40S subunits (Fig. 7D, lanes 2,5,11) than at these positions on wild-type CSFV RNA (Fig. 3D, lane 2) or on the UC 325 GA 357 pseudorevertant (Fig. 7D, lane 8) that had been incubated in identical conditions. After incubation with eif2, eif3, 40S ribosomal subunits, Met trna Met i, and GMP PNP, RT arrest at UGA was relatively weaker on UC 325,GA 357, and A 349 A 353 mutant RNAs (Fig. 7D, lanes 3,6,12) than on wild-type CSFV RNA (Fig. 3D, lane 5) or on the UC 325 GA 357 pseudorevertant (Fig. 7D, lane 9). A prominent stop caused by bound eif3 appeared at A 250 on all four mutant RNAs (Fig. 7D, lanes 3,6,9,12). These results indicate that the 3 half of the CSFV pseudoknot is not a primary determinant of 40S subunit binding to this IRES but that it influences an interaction 76 GENES & DEVELOPMENT

11 Internal translation initiation between them that may be required for correct positioning of the initiation codon in the ribosomal P site and therefore for stable assembly of 48S ribosomal complexes at the CSFV initiation codon. Taken together with the results of analysis of the HCV deletion mutants, these observations indicate that 48S complex assembly at the initiation codon is determined by RNA elements that are dispersed throughout the linear IRES sequence. Deletion of nucleotides abrogates binding of 40S subunits to the CSFV IRES 40S subunits bound much less strongly to CSFV nucleotides than to CSFV nucleotides RNA (Fig. 2I). This result indicates that an important determinant of IRES 40S subunit binding is located near the base of domain III. We deleted nucleotides from the basal helix of this domain (Fig. 7C) to determine whether its integrity is required for binary complex formation. CSFV IRES-mediated translation of the NS cistron was abolished by this deletion (Fig. 8A). The difference in translation of wild-type and mutant CSFV mrnas was not attributable to different stabilities of these templates in RRL. No significant difference in the degradation of these two templates was found after 15 min of incubation in RRL (data not shown). Ribosomal subunits (40S) bound wild-type [ 32 P]UTP-labeled CSFV nucleotides RNA stably; their binding to corresponding nucleotides 1 442( A 145 U 148 ) mutant RNA was abolished almost totally (Fig. 2J). This result was supported strongly by the results of toeprinting analysis. Primer extension was arrested by 40S subunits at C 334,G 345, and UUU on wild-type CSFV RNA (Fig. 8B, lane 2), whereas RT arrest at these positions on nucleotides 1 442( A 145 U 148 ) mutant RNA was insignificant (Fig. 8B, lane 5). A strong stop was detected at A 250 on CSFV nucleotides 1 442( A 145 U 148 ) mutant RNA incubated with GMP PNP, Met trna Met i, eif2, eif3, and 40S subunits (Fig. 8B, lane 5). This stop is caused by binding of eif3 to the CSFV IRES, which is therefore not affected by deletion of nucleotides In the same reaction, toeprinting did not yield stops at UGA This result indicates that 48S complexes are unable to form at the initiation codon on the nucleotides 1 442( A 145 U 148 ) mutant IRES. The pattern of endogenous strong stops on wildtype and nucleotides 1 442( A 145 U 148 ) mutant CSFV IRESs were similar, indicating that their structures did not differ grossly (Fig. 8B, lanes 1,4). cdnas shown in Figure 8B were uniformly labeled with [ - 32 P]dATP and their intensity is therefore proportional to their length, whereas cdnas shown in all other figures were endlabeled with [ - 32 P]ATP and their intensities are independent of length. Taken together, these results indicate that the nucleotides 1 442( A 145 U 148 ) CSFV mutant has a primary defect in binding to 40S subunits that is responsible for its inability to support assembly of 48S complexes and therefore to act as a template for translation in vitro. Figure 8. Effect of deleting nucleotides on CSFV IRESmediated translation, binding of 40S subunits and assembly of 48S complexes. (A) RRL (10 µl) was incubated alone (lane 1), with 0.5 µg wild-type CSFV (nucleotides 1 442) NS RNA (lane 2), or with 0.5 µg mutant CSFV (nucleotides ) RNA (lane 3) for 60 min at 30 C as described in Materials and Methods. Translation products were analyzed by autoradiography after electrophoresis on SDS-17% polyacrylamide gel. (B) Wild-type CSFV (nucleotides 1 442) NS (lanes 1 3) and mutant CSFV (nucleotides ) RNAs (lanes 4 6) were incubated with 40S subunits (lanes 2,3,5,6) and with eif2, eif3, Met trna Met i and GMP PNP (lanes 3,6). Reaction conditions are described in Materials and Methods. An unlabeled primer (5 -GGGATTTCTGATCTCGGCG-3 ) was annealed to the NS reporter cistron and was extended with AMV RT in the presence of [ - 32 P]dATP. RT stop sites are indicated on the right. Reference lanes C,T, A, and G depict the CSFV sequence. Ribosomal protein S9 interacts with HCV and CSFV IRESs in binary IRES 40S subunit complexes Primer extension analyses indicate that binary complexes described here are stabilized by multiple IRES 40S subunit interactions. We used UV cross-linking to determine whether specific ribosomal proteins bound to [ 32 P]UTP-labeled HCV RNA. A single 25-kD protein (p25) was labeled after UV cross-linking binary complexes of rabbit 40S subunits and [ 32 P]UTP-labeled HCV nucleotides RNA (Fig. 9A, lane 2). The same protein was one of several bands that appeared after UV cross-linking binary complexes of 40S subunits and [ 32 P]UTP-labeled CSFV nucleotides RNA (Fig. 9B, lane 3). These additional bands also appeared after UV cross-linking CSFV nucleotides ( A 145 U 148 ) RNA and 40S subunits, which do not form binary complexes (Fig. 9B, lane 4). They did not co-migrate with GENES & DEVELOPMENT 77

12 Pestova et al. Figure 9. UV cross-linking of ribosomal protein S9 to HCV and CSFV IRESs. (A) Proteins from rabbit ribosomal 40S subunits (lane 1) or from binary complexes of rabbit 40S ribosomal subunits and [ 32 P]UTP-labeled HCV nucleotides RNA (lane 2) were resolved by gel electrophoresis directly (lane 1) or after UV cross-linking and RNase digestion (lane 2). The positions of molecular mass marker proteins are indicated to the left of lane 1. (B) Proteins from binary complexes of rabbit 40S ribosomal subunits and [ 32 P]UTP-labeled HCV nucleotides RNA (lane 1), HCV nucleotides RNA (lane 2), CSFV nucleotides RNA (lane 3), or CSFV nucleotides 1 442( nucleotides ) (lane 4) were resolved by gel electrophoresis after UV cross-linking, and RNase digestion. (C) Proteins from binary complexes of rabbit 40S ribosomal subunits and [ 32 P]UTP-labeled HCV nucleotides RNA (lane 1), HCV nucleotides RNA (lane 2), HCV nucleotides RNA (lane 3), HCV nucleotides (lane 4), CSFV nucleotides (TC 325 GA 357 ) (lane 5), CSFV nucleotides ( A 349 A 353 ) (lane 6), CSFV nucleotides (GA 357 ) (lane 7), or CSFV nucleotides (TC 325 ) (lane 8) were resolved by gel electrophoresis after UV cross-linking, and RNase digestion. Proteins in A (lane 1) were visualized by staining with Coomassie blue; radio-labeled proteins in all other lanes in A,B, and C were visualized by autoradiography. ribosomal proteins and are probably products of incomplete RNase digestion of UV cross-linked CSFV RNA. p25 was not detected when [ 32 P]UTP-labeled HCV nucleotides RNA was cross-linked to 40S subunits (Fig. 9B, lane 2) although this RNA forms a stable binary complex with 40S subunits (Fig. 2F). p25 is therefore not a primary determinant of binary complex formation. p25 was also not detected when [ 32 P]UTP-labeled CSFV nucleotides ( A 145 U 148 ) RNA was crosslinked to 40S subunits (Fig. 9B, lane 4). This mutant RNA does not form a binary complex (Fig. 1J). The four nucleotides that are deleted in this CSFV RNA are part of a helix that is intact in HCV nucleotides RNA. Nevertheless, these data do not indicate whether p25 makes independent contact with these IRESs or whether such contact depends on proper positioning of the RNA determined by a prior primary recognition step. CSFV nucleotides are clearly required for primary recognition of the IRES by 40S subunits, but the effect of deleting these residues on structures elsewhere in the IRES that could determine potential independent interaction with p25 is not known. UV cross-linking of p25 was sensitive to changes in structure of different regions of these IRESs. Labeling of p25 after UV cross-linking of binary complexes assembled on wild-type and mutant HCV IRESs was reduced by deletion of hairpins IIa and IIIc but was not significantly affected by deletion of hairpin IIIb (Fig. 9C, lanes 1 4). Mutations within the CSFV pseudoknot also affected UV cross-linking of p25. p25 became labeled much less strongly after 40S subunits were cross-linked to [ 32 P]UTP-labeled GA 357 and A 349 A 353 mutant CSFV nucleotides RNAs than to corresponding UC 325 and UC 325 GA 357 CSFV RNAs (Fig. 9C, lanes 5 8). The identity of p25 was established by amino-terminal sequence analysis. The first 10 amino-acid residues of p25 were PVATRSWY(X)RK, in which (X) represents a residue whose identity could not be determined. Cysteine residues are destroyed chemically in the sequencing reaction. These residues match the amino terminus of the S9 ribosomal protein exactly (Chan et al. 1993). Discussion We have reconstituted internal ribosomal entry on HCV and CSFV IRESs in vitro to the stage of 48S complex formation using purified translation components to identify the necessary factors and to characterize their role and that of these two RNAs in this process. Our results show that the initiation mechanism used by HCV and CSFV IRESs differs from modes of eukaryotic initiation described previously. Some of the unusual properties of HCV and CSFV IRES-mediated initiation have parallels in the process of translation initiation in prokaryotes. Factors and cofactors required for internal ribosomal entry Ribosomal subunits (40S) formed stable binary complexes on HCV and CSFV IRESs in which the initiation codon is in the ribosomal P site. The processes of ribosomal binding to these IRESs, positioning of the ribosome at the initiation codon, and formation of active 80S complexes did not require the initiation factors eif4a, eif4b, and eif4f and were not influenced by them. Initiation mediated by these two IRESs therefore differs fundamentally from initiation on all other eukaryotic mrnas, including even internal initiation mediated by the EMCV IRES, which depends absolutely on one or more of these factors (Trachsel et al. 1977; Benne and Hershey 1978; Blum et al. 1989; Pause et al. 1994; Pestova et al. 1996a,b). This unexpected observation was 78 GENES & DEVELOPMENT